Forecast Error Metrics: MAPE, WAPE, MAE, and RMSE

WAPE (weighted absolute percentage error) is total absolute forecast error divided by total actual volume, so busy intervals count more than quiet ones. MAPE, MAE and RMSE measure error in other ways, and each can make the same forecast look better or worse. This guide explains which metric to trust for staffing decisions.

Want forecasts you can measure this way without spreadsheets? See workload forecasting in Soon, and how forecasts feed auto scheduling software.

That matters because workforce teams do not evaluate forecasts for academic reasons. They evaluate them because a forecast drives staffing assumptions, scheduling decisions, and coverage risk. If the metric is misleading, the planning decision can be misleading too.

This is where teams often get stuck. One forecast might look good on MAPE and weak on RMSE. Another might have strong average accuracy but fail badly in the busiest intervals. A third might produce clean-looking point forecasts but poorly calibrated uncertainty ranges. None of those are small details if the forecast is being used to decide coverage.

A better question is not "Which metric is best?" It is which metric helps us understand whether this forecast is good enough for the decisions we need to make.

This guide explains the most common forecast error metrics, where each one is useful, where each one can mislead you, and which combinations tend to work best for workforce planning teams.

What makes a forecast metric useful

A useful metric does more than summarize model performance. It reflects the kind of mistakes that matter operationally.

• It should fit the shape of the data, including low-volume periods, spikes, and uneven demand.
• It should help planners judge whether the forecast is reliable enough to staff against.
• It should not hide the kind of misses that are expensive in practice, such as large errors in high-volume intervals.
• It should be interpretable enough that teams can actually use it in planning reviews.

MAPE (Mean Absolute Percentage Error)

MAPE tells you, on average, how far off the forecast was in percentage terms. That is why people like it. It is easy to explain and easy to compare at a glance.

If a planner says a forecast was off by 8%, most people immediately understand what that means.

Where MAPE helps

• quick percentage-based reporting
• simple comparisons when volumes are reasonably stable
• stakeholder conversations where a relative error is easier to understand than an absolute one

Where MAPE misleads

• low-volume or zero-volume intervals can make the metric unstable or unusable
• small absolute misses can look dramatic when the denominator is tiny
• it can make a forecast look worse than it feels operationally in sparse demand patterns

For workforce teams, this matters because many real operations do have quiet intervals. If your volume dips close to zero overnight or between peaks, MAPE can distort the picture.

WAPE (Weighted Absolute Percentage Error)

WAPE is often more practical in operational forecasting because it puts the error in relation to total actual volume instead of averaging each percentage equally.

That makes it more stable in environments where some intervals matter much more than others.

Where WAPE helps

• operations with uneven demand across the day or week
• cases where high-volume periods should influence the metric more than quiet ones
• forecast reviews where planners want a more operationally grounded percentage view

What WAPE still misses

• it can still hide when the largest misses happen in the most sensitive intervals
• it does not tell you whether a few large misses are driving the problem
• it is still one summary number, not a full picture of forecast risk

For many workforce teams, WAPE is a better overall accuracy metric than MAPE. But it still should not be the only thing you track.

MAE (Mean Absolute Error)

MAE tells you the average size of the miss in the original unit, such as contacts, tickets, or hours. That is useful because it is often closer to how planners actually think.

If your average miss is 18 contacts per interval, that can be easier to translate into staffing implications than a percentage alone.

Where MAE helps

• understanding average miss size in operational units
• comparing forecast models when you want a straightforward average error
• staffing conversations where absolute error is easier to connect to coverage impact

Where MAE is limited

• it does not show whether the miss is large or small relative to demand
• the same absolute miss can be trivial in one interval and serious in another
• it does not especially emphasize the biggest misses

RMSE (Root Mean Square Error)

RMSE is useful when large misses are especially painful. Because it squares the errors before averaging them, it gives more weight to bigger mistakes.

That makes it useful in planning environments where a few large misses can do more damage than many small ones.

Where RMSE helps

• surfacing forecasts that occasionally fail badly
• operations where large misses create outsized coverage problems
• model comparisons where tail risk matters, not just average miss size

What to watch with RMSE

• it is sensitive to outliers
• it can make a forecast look poor because of a handful of extreme misses
• it needs interpretation alongside other metrics, not in isolation

For workforce planning, RMSE becomes more valuable when the cost of a large miss is operationally severe, for example when understaffing a peak creates service failure or backlog that spills into later intervals.

Coverage metrics for uncertainty ranges

Point forecasts are not the full story. Good forecasting systems should also help teams understand uncertainty. Coverage metrics do that by testing whether forecast ranges are calibrated well enough.

If you say the actual result should fall inside a prediction interval 90% of the time, coverage tells you whether that is actually true.

Why coverage matters

• it shows whether a model is overconfident or too cautious
• it helps planners judge whether ranges are useful for decision-making
• it is important when staffing decisions depend on risk, not just the midpoint forecast

What coverage does not solve on its own

• very wide intervals can look well calibrated while still being too vague to use
• good coverage does not mean the point forecast is strong
• you still need to ask whether the range is actionable for staffing decisions

Which metrics should a workforce team actually use

Usually not one. Most teams benefit from using a small combination of metrics because each one highlights a different kind of forecasting problem.

1. Use WAPE for a stable overall relative view of error.
2. Use MAE or RMSE to understand absolute miss size and whether large misses are hurting you.
3. Use coverage metrics when forecast ranges are part of the planning process.

The exact combination depends on how the forecast is used, but one metric alone is rarely enough if the forecast is meant to support real staffing decisions.

What is a good WAPE or MAPE?

There is no single good number. Published benchmarks for contact volume accuracy are rare, and the ones that circulate rarely say which interval, horizon, or queue size they were measured on. A WAPE that is excellent for one operation can be poor for another. What you can do is understand what moves the number and set a target against your own baseline.

What drives the number

• Interval granularity. Error measured per 15-minute interval is almost always higher than error on the same forecast summed to the day or week, because random noise partly cancels out when intervals are added together. Always state the granularity next to the metric.
• Volume per interval. Small queues have proportionally more noise. MAPE in particular becomes unstable or undefined when actual volume is close to zero, which Hyndman and Athanasopoulos list as a core weakness of percentage errors.
• Forecast horizon. A forecast made one day out is usually more accurate than one made six weeks out. Compare like with like.
• Events and changes. Launches, outages, campaigns, and routing changes create misses no model could see in history. Tag them so they do not hide the model's normal performance.
• Channel mix. Volume on newer or campaign-driven channels tends to be noisier than an established phone queue.

A practical target: beat a simple benchmark

Instead of chasing an external number, compare your forecast with a simple benchmark on the same data. A seasonal naive forecast (the same interval last week) is a common choice, and Forecasting: Principles and Practice recommends checking any new method against such simple methods first. If your model does not clearly beat seasonal naive in the intervals that drive staffing, it is not ready to plan with.

Backtest walkthrough

1. Pick a history window that includes your normal seasonality, ideally a full year, and choose the granularity you staff at, such as 15 or 30 minutes.
2. Hold out the most recent weeks as a test set that the model never sees during fitting.
3. Use a rolling origin: forecast the next period, move the cut-off forward, and repeat, the approach described as time series cross-validation.
4. For every origin, calculate WAPE for the whole period, MAE in contacts per interval, and the same metrics for a seasonal naive benchmark.
5. Break the error down by interval of day and day of week. Large misses in the peak intervals matter more than a good daily average.
6. Check bias separately. A forecast that is consistently 5% low is easier to fix, and more dangerous to ignore, than one that is noisy around zero.
7. Translate the result into staffing. Run the forecast and the actuals through the same staffing calculation and see whether the error would have changed the number of agents scheduled.

You can try this on your own data with the free Soon forecast tool, which runs in the browser on an uploaded CSV. Forecasting inside the Soon app, including backtests and uncertainty bands that feed staffing, is part of the Scale plan.

The bigger mistake teams make

The biggest mistake is treating forecast accuracy as a model problem instead of a staffing problem.

• Teams look at averages only and miss where the painful intervals are.
• They review overall error but ignore whether the forecast failed during the busiest windows.
• They compare models without asking whether the errors would have changed staffing decisions.
• They treat a cleaner dashboard as success, even if coverage outcomes did not improve.

This is also why forecast quality connects directly to broader workforce planning decisions. Once a team trusts the forecast, it can make better use of staffing logic such as Erlang models and simulation.

Good metrics help teams make better decisions

The best forecast metric depends on how the forecast is being used. A simple percentage metric may be enough for high-level reporting. A planning team making staffing calls usually needs something more grounded.

What matters most is not whether the metric looks sophisticated. It is whether it helps the team decide if the forecast is good enough to act on.

That is where Soon forecasting fits. Better forecasting workflows make it easier to evaluate forecast quality, understand where the real risk is, and turn that insight into stronger staffing decisions.